allenai_Olmo-3.1-32B-Think GGUF size and VRAM requirements

License: apache-2.0 ⬇ 14,476 ❤ 5
Parameters32.23B
Context65,536

bartowski/allenai_Olmo-3.1-32B-Think-GGUF is a very large language model with 32.23 billion parameters, built on the olmo2 architecture. It is released under the apache-2.0 license and has been downloaded 14,476 times.

Under the hood it uses 64 transformer layers, a hidden size of 5,120, 40 attention heads. It uses a hybrid attention design: only 16 of its 64 layers keep a full KV cache, while the rest use linear attention with a constant-size state. Long contexts therefore cost far less memory than on a conventional model of this size.

To run bartowski/allenai_Olmo-3.1-32B-Think-GGUF locally at a 4,096-token context, its quantized versions need between 9.2 GB (IQ2_XXS, lowest quality) and 61.09 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included. Context length drives that memory directly: at 4,096 tokens the KV cache for allenai_Olmo-3.1-32B-Think-GGUF is about 0.25 GB, rising to roughly 4.0 GB at its full 65,536-token context. Shorter prompts free up memory for a higher-quality quantization.

For most users the best balance is Q5_K_L, needing about 22.63 GB. That means bartowski/allenai_Olmo-3.1-32B-Think-GGUF fits entirely in the VRAM of a 10 GB GPU or larger, running fully on the GPU.

Available GGUF quantizations for bartowski/allenai_Olmo-3.1-32B-Think-GGUF include IQ2_XXS, IQ2_XS, IQ2_S, IQ2_M, Q2_K, Q2_K_L, IQ3_XXS, IQ3_XS, Q3_K_S, IQ3_M, Q3_K_M, Q3_K_L, IQ4_XS, Q3_K_XL, IQ4_NL, Q4_0, Q4_K_S, Q4_K_M, Q4_K_L, Q4_1, Q5_K_S, Q5_K_M, Q5_K_L, Q6_K, Q6_K_L, Q8_0, BF16. The model supports a native context length of up to 65,536 tokens; a longer context grows the KV cache and the memory needed.

→ Guide: How much VRAM do you need?

GGUF file size and memory by quantization

Compare real GGUF weight sizes, estimated KV cache and total memory for Q4, Q5, Q8 and every quantization published in this repository.

Quant.Bits QualityWeights KVTotal Speed~Verdict
IQ2_XXS 2.17 Very low 8.15 GB 0.25 GB 9.2 GB 6.1 t/s Offload
IQ2_XS 2.4 Very low 9.02 GB 0.25 GB 10.07 GB 5.5 t/s Offload
IQ2_S 2.5 Very low 9.4 GB 0.25 GB 10.45 GB 5.3 t/s Offload
IQ2_M 2.72 Low 10.21 GB 0.25 GB 11.26 GB 4.9 t/s Offload
Q2_K 2.98 Low 11.18 GB 0.25 GB 12.23 GB 4.5 t/s Offload
Q2_K_L 3.1 Low 11.65 GB 0.25 GB 12.7 GB 4.3 t/s Offload
IQ3_XXS 3.11 Low 11.68 GB 0.25 GB 12.73 GB 4.3 t/s Offload
IQ3_XS 3.32 Fair 12.45 GB 0.25 GB 13.5 GB 4.0 t/s Offload
Q3_K_S 3.49 Fair 13.09 GB 0.25 GB 14.14 GB 3.8 t/s Offload
IQ3_M 3.59 Fair 13.48 GB 0.25 GB 14.53 GB 3.7 t/s Offload
Q3_K_M 3.87 Fair 14.53 GB 0.25 GB 15.58 GB 3.4 t/s Offload
Q3_K_L 4.2 Fair 15.75 GB 0.25 GB 16.8 GB 3.2 t/s Offload
IQ4_XS 4.3 Good 16.14 GB 0.25 GB 17.19 GB 3.1 t/s Offload
Q3_K_XL 4.31 Good 16.17 GB 0.25 GB 17.22 GB 3.1 t/s Offload
IQ4_NL 4.55 Good 17.06 GB 0.25 GB 18.11 GB 2.9 t/s Offload
Q4_0 4.55 Good 17.08 GB 0.25 GB 18.13 GB 2.9 t/s Offload
Q4_K_S 4.57 Good 17.15 GB 0.25 GB 18.2 GB 2.9 t/s Offload
Q4_K_M 4.84 Good 18.14 GB 0.25 GB 19.19 GB 2.8 t/s Offload
Q4_K_L 4.93 Good 18.5 GB 0.25 GB 19.55 GB 2.7 t/s Offload
Q4_1 5.03 Very good 18.86 GB 0.25 GB 19.91 GB 2.7 t/s Offload
Q5_K_S 5.52 Very good 20.71 GB 0.25 GB 21.76 GB 2.4 t/s Offload
Q5_K_M 5.67 Very good 21.29 GB 0.25 GB 22.34 GB 2.3 t/s Offload
Q5_K_L 5.75 Very good 21.58 GB 0.25 GB 22.63 GB 2.3 t/s Offload
Q6_K 6.56 Excellent 24.63 GB 0.25 GB 25.68 GB Insufficient
Q6_K_L 6.63 Excellent 24.86 GB 0.25 GB 25.91 GB Insufficient
Q8_0 8.5 Excellent 31.9 GB 0.25 GB 32.95 GB Insufficient
BF16 16.0 Excellent 60.04 GB 0.25 GB 61.09 GB Insufficient

KV cache computed from the model's exact architecture. Speed is a rough estimate bounded by memory bandwidth.

Frequently asked questions

What kind of model is bartowski/allenai_Olmo-3.1-32B-Think-GGUF?

bartowski/allenai_Olmo-3.1-32B-Think-GGUF is a language model with 32.23 billion parameters, based on the olmo2 architecture. It is released under the apache-2.0 license and distributed as GGUF files for local inference.

How does bartowski/allenai_Olmo-3.1-32B-Think-GGUF handle long context?

bartowski/allenai_Olmo-3.1-32B-Think-GGUF uses a hybrid attention design: only 16 of its 64 layers keep a full KV cache, and the rest use linear attention with a constant-size state. Long contexts cost far less memory than on a conventional model of this size.

How much VRAM do you need to run bartowski/allenai_Olmo-3.1-32B-Think-GGUF?

You need about 9.2 GB of VRAM to run bartowski/allenai_Olmo-3.1-32B-Think-GGUF entirely on the GPU using the IQ2_XXS quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.

Can I run bartowski/allenai_Olmo-3.1-32B-Think-GGUF on an 8 GB GPU?

Partially. bartowski/allenai_Olmo-3.1-32B-Think-GGUF only fits on an 8 GB GPU by offloading part of it to system RAM (with Q5_K_L), which runs but is slower.

Can I run bartowski/allenai_Olmo-3.1-32B-Think-GGUF on a 16 GB GPU?

Yes. With 16 GB of VRAM you can run bartowski/allenai_Olmo-3.1-32B-Think-GGUF fully on the GPU using Q3_K_M (about 15.58 GB).

Can I run bartowski/allenai_Olmo-3.1-32B-Think-GGUF on a 24 GB GPU?

Yes. With 24 GB of VRAM you can run bartowski/allenai_Olmo-3.1-32B-Think-GGUF fully on the GPU using Q5_K_L (about 22.63 GB).

What context length does bartowski/allenai_Olmo-3.1-32B-Think-GGUF support?

bartowski/allenai_Olmo-3.1-32B-Think-GGUF supports a native context length of up to 65,536 tokens. A longer context grows the KV cache, so it increases the memory needed to run the model.

What is the best quantization for bartowski/allenai_Olmo-3.1-32B-Think-GGUF?

For bartowski/allenai_Olmo-3.1-32B-Think-GGUF, a strong default is Q4_K_M, which needs about 19.19 GB and keeps most of the quality while roughly halving the memory versus 8-bit. With VRAM to spare, Q5_K_M or Q6_K add a little more quality; if you are tight on memory, a smaller quantization still runs. Pick the highest quantization that fits your VRAM.